What is a Dataset?

What it is: A dataset is a big collection of information used to teach an AI. The AI studies the dataset until it learns patterns.
Who it’s for: Beginners learning how AI is trained
Best if: You want to know where AI ‘knowledge’ comes from
Skip if: You build datasets for a living

A dataset is what the AI learns from. It’s a pile of information. Lots of it.

Think of it like a cookbook for someone learning to cook. The more recipes in the book, the more dishes they can make. If the cookbook only has pasta recipes, they can only cook pasta. If the cookbook has recipes from every country, they can cook almost anything.

What goes into an AI dataset

It depends on what the AI is learning.

• For a writing AI: Books, articles, web pages, Wikipedia, conversations.
• For an image AI: Millions of pictures with labels (“this is a cat,” “this is a sunset”).
• For a voice AI: Hours of recorded speech with the words written out.
• For a self-driving car AI: Video of roads, traffic, and weather.

Why dataset quality matters

If the dataset has bad info, the AI learns bad info. That’s where AI bias comes from. If an AI only sees pictures of doctors who are men, it might think doctors are always men. The fix is better datasets.

Rule: garbage in, garbage out. Good datasets make good AI.

You May Also Like

• What is Training Data?
• What is Bias in AI?
• What is Machine Learning?
• AI Glossary: 100+ Terms Every Beginner Needs to Know